(02-26-2015, 07:31 AM)NaturalViolence Wrote: [ -> ]Please redo the test with efb to ram and LLE audio. Which build was this on?
4.0 - 5662. Well with efb to ram and lle audio i am getting with 4x similar fps i was getting with efb to texture and hle with 6x (about 83 fps). 72 fps using 5x ir and 63 fps using 6x ir which makes it barely playable. With those settings though you get both a cpu and gpu bottleneck.
Was this with openGL or d3d11? Per-pixel or per-vertex lighting? Please retest with both video backends.
Now I need someone to test with an AMD gpu that gets comparable performance in PC games like the 280/285 or even a 270X. And then I need someone to test with an older nvidia gpu that gets similar performance in PC games but has more memory bandwidth. The GTX 760 would be the best choice. Then and only then will we have the data needed to arrive at a safe conclusion about this "issue".
(02-26-2015, 12:17 PM)NaturalViolence Wrote: [ -> ]Was this with openGL or d3d11? Per-pixel or per-vertex lighting? Please retest with both video backends.
Now I need someone to test with an AMD gpu that gets comparable performance in PC games like the 280/285 or even a 270X. And then I need someone to test with an older nvidia gpu that gets similar performance in PC games but has more memory bandwidth. The GTX 760 would be the best choice. Then and only then will we have the data needed to arrive at a safe conclusion about this "issue".
D3D and per pixel is disabled. Jmc has a 760 i think?
Test in openGL too please.
(02-26-2015, 11:05 PM)NaturalViolence Wrote: [ -> ]Test in openGL too please.
I am getting pretty bad framerates with opengl and i don't know why. Same settings as d3d 4x ir gives 105 fps, 5x gives 80 fps and 6x gives 62 fps, and those were using efb to texture and hle audio. With efb to ram and lle things are even worse 75 fps at 4x, 60 fps at 5x and 50 fps at 6x. Perhaps it is a driver issue or something...
UPDATE: Finally built my computer like 2 weeks ago and so far, I'm pretty happy with the performance, if anyone needs to do some tests or anything, feel free to let me know.
@LttP:
For accurate (GPU) benchmark results,
disable CPU->EFB Access and use a 100% complete (241s) save file.
For easy benchmarking, use Dolphin-Ishiiruka (with up to 8xIR selectable from the UI) or compile your own master build with
PR #2024.
kirbypuff Wrote:and use a 100% complete (241s) save file.
Why would this have any effect at all?
kirbypuff Wrote:For accurate (GPU) benchmark results, disable CPU->EFB Access
kirbypuff Wrote:For easy benchmarking, use Dolphin-Ishiiruka (with up to 8xIR selectable from the UI) or compile your own master build with PR #2024.
This defeats the point. We're trying to asses how these cards perform with current dolphin builds under ordinary scenarios.
(03-02-2015, 02:21 AM)NaturalViolence Wrote: [ -> ]Why would this have any effect at all?
As you complete more, that particular scene gets more demanding (more object/models added to the scene).
(03-02-2015, 02:21 AM)NaturalViolence Wrote: [ -> ]This defeats the point. We're trying to asses how these cards perform with current dolphin builds under ordinary scenarios.
It's much faster to increase / decrease the IR in real-time (until the GPU becomes the bottleneck) than to stop and restart Dolphin every time the IR is changed.
The current master build can only go up to 4x from the UI and that's not enough.